Study On Key Technologies In Non-intrusive Load Disaggregation For Intelligent Power Utilization

Load Monitoring technology belongs to the field of advanced measurement infrastructure of smart grid,and is used to obtain real-time energy consumption information of electrical appliances in power users’ homes.Load monitoring technology can provide data support for refined and customized services such as energy efficiency management,power demand-side management and time-of-use tariff formulation of smart grid,which is of great research significance.At present,obtaining detailed energy consumption information of a user’s internal appliances is mainly achieved by installing sensors on each appliance,which will bring privacy and security issues to users.In addition,the high hardware maintenance and installation costs further increase the economic costs,making it difficult for the technology to be rolled out on a large scale among power users.Non-intrusive load disaggregation technology can obtain real-time electricity consumption information of each appliance by constructing the corresponding calculation model to analyze the aggregated load information.It has the advantages of easy acceptance,easy implementation and low cost.However,the current non-intrusive load disaggregation technology has problems such as data acquisition difficulties,low efficiency of load disaggregation methods and low accuracy of disaggregation results.In view of the above problems,this paper conducts research from both hardware and software aspects,combines the hidden Markov model and deep learning model to propose a new load disaggregation method and develop a load disaggregation device,which provides a theoretical basis and hardware reference for the large-scale promotion and application of non-intrusive load disaggregation technology for intelligent power utilization,and the specific research content includes:(1)For the problem that the traditional Hidden Markov Model(HMM)ignores the accurate analysis of the operation state of appliance,resulting in the low disaggregation accuracy of multi-state appliance,an adaptive Factorial Hidden Markov Model(FHMM)is proposed.The method uses an adaptive density peak clustering method to analyze the power consumption sequences of appliance,so as to automatically determines the operation states of different appliances and its corresponding power consumptions,and improve the disaggregation accuracy of the model.On this basis,combined with the FHMM model,a complex combination chain in the traditional HMM model is reduced to multiple low-dimensional simple chains,which improves the computational efficiency of the model.(2)For the problem that the traditional load disaggregation model only uses a single electrical feature for modeling,and it is difficult to effectively distinguish the electrical appliance with similar characteristics,and an efficient hidden semi-Markov model based on the duration of electrical appliance operation state is proposed.Through the analysis of the traditional HMM model,it can be seen that the duration distribution of the appliance is similar to the geometric distribution,which limits the practicality of the HMM model.The paper proposes to use the Poisson distribution to characterize the duration distribution of the operation state of electrical appliance.Combined with the power sequence of appliance,the disaggregation accuracy of the model is effectively improved.On the other hand,the computation time of the model is greatly reduced by reducing the constraints of the training algorithm and converting the matrix multiplication operation into an additive operation.(3)For the characteristics of load space dynamic change in non-intrusive load disaggregation technology,a disaggregation method for unknown loads is proposed.This method uses the improved Power Derivative Fast DTW(PD-Fast DTW)algorithm to measure the similarity between unknown loads and known loads,solves the singularity problem of the traditional DTW method and improves the computational efficiency of the algorithm.In addition,in order to improve the adaptability and generalizability of non-intrusive load disaggregation technology in different power consumption scenarios,a load disaggregation model(Im2Seq)based on image features is proposed.Through the adaptive state clustering method,the operation state of the appliance is accurately described,and the usage time of the appliance is introduced into the modeling process,so that the one-dimensional power time series is converted into a two-dimensional image containing the operation state conversion information of appliance and temporal information,which effectively enhances the feature differences of different devices.Finally,the PD-Fast DTW method is combined with the load disaggregation method Im2 Seq to realize the disaggregation of unknown loads.(4)For the requirements of non-intrusive load disaggregation technology for data sampling rate and real-time disaggregation,a system that can realize real-time load disaggregation is designed.The system is mainly divided into two parts: the cloud platform and the load disaggregation device.The cloud platform embeds a comprehensive load disaggregation method to process large-scale electricity consumption data.The load disaggregation device collects power consumption data of different frequencies according to the demand,and embeds the real-time load disaggregation method for real-time load disaggregation.In addition,the device can locally store and model the collected data,thereby effectively protecting the privacy of users.